Could Shopping Data Be Used to Predict Cancer?

A recent study suggested that shopping data collected through credit cards could help predict cancer risk. Researchers are now examining how to use this information and what it might mean for patients.

The study found that patients with ovarian cancer bought more digestive medicines and painkillers in the months before diagnosis than people without cancer. It used data that participants agreed to share from credit-card purchases at two UK retailers, a pharmacy and a supermarket.

Building on this research, investigators are starting a larger study to examine whether purchasing data are associated with diagnoses of 10 different cancers. If the initial findings can be replicated, the team will consider ways to inform consumers about whether their shopping habits are linked to cancer risk.

Experts caution, however, that identifying purchases predictive of cancer or other diseases may be very difficult. Their predictive power would need to be extremely high to justify notifying and potentially alarming consumers. Consumers and advocacy groups have also raised concerns about research use of shopping data, particularly whether it could be repurposed for marketing or other uses they have not authorized.

Shopping Data and Cancer Risk: Findings and Implications

The UK study, published in JMIR Public Health and Surveillance , included 153 women with ovarian cancer from clinics in England, Scotland and Wales, and 306 women of similar ages without ovarian cancer recruited online. All agreed to share up to 6 years of historical credit-card data from 2 retailers.

Researchers found a significant association between purchases of digestive medicines and painkillers and ovarian cancer diagnosis, with the strongest association 8 months before diagnosis (odds ratio [OR], 2.91; 95% CI, 2.07-4.12). Digestive medicines were the main driver; purchases began increasing 9 months before diagnosis (OR, 1.38; 95% CI 1.04-1.83).

The strongest association was among women diagnosed with advanced ovarian cancer. In that group, purchasing history could predict 68% of cases.

Patients with advanced or late-stage disease have a worse prognosis than those with early-stage disease, experience symptoms and are precisely the patients who need earlier diagnosis.

Early ovarian cancer diagnosis can be difficult because it tends not to cause symptoms initially. Symptoms generally include digestive complaints and stomach and back pain, which even doctors may miss.

The ultimate aim was to encourage earlier medical visits and shorten the interval between first symptoms and seeking help. However, purchasing patterns are not strongly enough associated with ovarian cancer.

Study author Dr. James M. Flanagan and colleagues are now looking more deeply into credit-card data to identify combinations of purchases that might strongly predict ovarian cancer risk, perhaps over-the-counter medicines combined with particular foods or vitamins.

Certain combinations might instead be more general signs of persistent bloating due to ovarian cancer, another cancer or conditions such as irritable bowel syndrome. Such a signal might justify informing consumers that they could have a digestive condition and may wish to discuss it with a doctor.

Using Personal Data: Consumer Perspectives and Data Protection Laws

Dr. Flanagan and colleagues conducted several studies, including face-to-face focus groups and telephone interviews, to explore consumers' views on sharing credit-card data for health research.

In one study, 54% were willing to share such data for research. Understanding the purpose of data donation increased willingness, but a sense of "social duty" was the strongest predictor of whether consumers would donate.

Consumers were particularly concerned about data management, use for purposes outside the study without their knowledge, and anonymization to protect identity and location.

Providing some of these safeguards may be harder than in the UK and Europe, where data protection laws give people ownership of and agency over personal information.

"[In the US], there is currently a patchwork of state and sector-specific laws that, in many cases, do a very poor job of governing how a [retail store] can use information collected through its credit program and what it must disclose initially," said John Davisson, senior counsel and director of litigation at the nonprofit research and advocacy organization EPIC.

He noted, however, that retailers differ from research studies. Studies generally undergo institutional review board oversight, which provides strict safeguards for secure data management. Even if a grocery store could freely give credit-card data to a third party such as a research group, researchers would probably still need consumers' consent.

Next Steps: More Research and Real-World Applications

Current research is far from providing sufficient evidence to implement any cancer warning system through credit-card programs.

The study was a proof of concept examining whether this personal information could be useful. If it proves useful, building systems that act on it would be a completely separate question.

If Dr. Flanagan and colleagues validate the initial findings, they hope to study the most acceptable ways to notify consumers. Interviews with researchers, patients and primary care physicians have identified possible approaches. For example, credit-card systems might send consumers or their doctors a letter about an algorithm-detected risk, or automatically arrange a diagnostic-center appointment.

It is also important to consider that a credit-card system using purchasing data to detect cancer risk in real-world settings could miss substantial population groups.

Using purchasing data for a real-world health notification system adds "another layer of complexity" compared with research. The same safeguards must govern who can access the data and how it may be used.

When consumers are interested in sharing data, great care is needed in explaining how their information is used and communicating the algorithm's results.

In conclusion, this new research offers important ideas for future applications of credit-card purchasing data to predict cancer risk. Ethical and legal challenges must also be considered. Addressing privacy concerns, explaining data use and obtaining consent are essential. Successful real-world applications will also require more research into the accuracy and reliability of algorithms predicting cancer risk from shopping data. Further work could advance the field and potentially contribute to early-detection and prevention methods that save lives.